#opendaylight-meeting: MD-SAL interest call

Meeting summary

1. agenda bashing (colindixon, 16:01:39)
   1. the webex didn’t have working audio, but now it seems to be back up and working correctly (colindixon, 16:04:19)
   2. https://wiki.opendaylight.org/view/MD-SAL_Weekly_Call#Agenda the agenda in its usual place (colindixon, 16:04:58)
   3. first-up is is Andrew McLachlan on the MD-SAL message bus (colindixon, 16:06:32)
   4. second-up is looking at Anton’s work on yangtools performance and address normalization in the MD-SAL to avoid bugs (colindixon, 16:07:40)
   
   
2. MD-SAL message bus (colindixon, 16:07:53)
   1. ACTION: Andrew and Tony to post slides after the call (colindixon, 16:08:57)
   2. (recording started just at the beginning of the presentation) (colindixon, 16:09:15)
   3. the key idea here is to provide publish/subscribe interfaces from ODL (colindixon, 16:10:38)
   4. current appraoch focuses on event sources and user agents (e.g., XMPP, 0MQ, etc.) (colindixon, 16:13:47)
   5. core functions are: (i) an event source manager, (ii) an event source topology/inventory, (iii) topic creation/subscription API (basically filtering) (colindixon, 16:14:56)
   6. colindixon asks why isn’t this notifications (which we have) and filters (which people have been desperately asking for forever)? (colindixon, 16:16:45)
   7. rovarga_ responds by saying that this is different because notifications are within the controller itself, and this is really about transporting events though the controller without modifying them (colindixon, 16:19:07)
   8. this is the classing node-centric/network-centric conflict (rovarga_, 16:20:42)
   9. it is essentially the same thing, but MD-SAL APIs are node-centric (e.g. single node, the controller itself), whereas this plugs 'the same thing' at the network layer (rovarga_, 16:21:28)
   10. anton, andrew and rovarga_ discuss how you may or may not need to normalize models and events to make it easy to use (colindixon, 16:28:37)
   11. tonly goes thorugh the event diagram (colindixon, 16:37:18)
   12. as far as I can tell, the key idea is two fold: (1) provide a way to push subscription to events/notifications outside the controller to devices and (2) a way to transport them outside the controller to various other “user-agents”, e.g., AQMP, 0MQ, etc. (colindixon, 16:42:40)
   13. there is a long debate about how different this is from the notifications we have already in YANG (colindixon, 16:54:27)
   14. rovarga_ is strongly of the opinion that they are *very* different (colindixon, 16:54:44)
   15. colindixon (and it seems a few others) seem less convinced that’s the case (colindixon, 16:55:12)

Action items

1. Andrew and Tony to post slides after the call

People present (lines said)

1. colindixon (40)
2. alagalah (31)
3. tbachman (20)
4. rovarga_ (11)
5. odl_meetbot (4)
6. phrobb (2)
7. devinavery (1)